Robust monitoring of hypovolemia in intensive care patients using photoplethysmogram signals

Abstract

The paper presents a fingertip photoplethysmography based technique to assess patient fluid status that is robust to waveform artifacts and health variability in the underlying patient population. The technique is intended for use in intensive care units, where patients are at risk for hypovolemia, and signal artifacts and inter-patient variations in health are common. Input signals are preprocessed to remove artifact, then a parameter-invariant statistic is calculated to remove effects of patient-specific physiology. Patient data from the Physionet MIMICII database was used to evaluate the performance of this technique. The proposed method was able to detect hypovolemia within 24 hours of onset in all hypovolemic patients tested, while producing minimal false alarms over non-hypovolemic patients. Abstract-The paper presents a fingertip photoplethysmography based technique to assess patient fluid status that is robust to waveform artifacts and health variability in the underlying patient population. The technique is intended for use in intensive care units, where patients are at risk for hypovolemia, and signal artifacts and inter-patient variations in health are common. Input signals are preprocessed to remove artifact, then a parameter-invariant statistic is calculated to remove effects of patient-specific physiology. Patient data from the Physionet MIMICII database was used to evaluate the performance of this technique. The proposed method was able to detect hypovolemia within 24 hours of onset in all hypovolemic patients tested, while producing minimal false alarms over non-hypovolemic patients.

Parameter-Invariant Monitor Design for Cyber–Physical Systems

“…Theoretically, PAIN designs can achieve a constant false alarm rate (CFAR) regardless of parameter uncertainty. Practically, PAIN designs have provided nearconstant false alarm rates in many CPS applications spanning networks [80], smart buildings [91], and medicine [92], [82], [83].…”

“…In this model, there are no restrictions on what the parameters represent physically. For instance, the parameters can represent physical parameters (e.g., metabolic rate [82]), but could also represent lumped parameters that have no explicit physical world interpretation (e.g., coefficients of a transfer function [83], [91]). In PAIN monitoring, we assume that µ ∈ Γ µ = R |F |c , ρ i ∈ Γ ρ,i = {ρ ∈ R |Gi|c | ρ = 1}, and σ ∈ Γ σ = {σ ∈ R | σ > 0} are nuisance parameters, and θ 0 , θ 1 ∈ R ≥0 are test parameters.…”

“…An alternative approach shown to be robust to data sparsity and inter/intra-system variability is parameter-invariant (PAIN) monitoring [79], [80], [81], [82], [83], [84], [85]. In PAIN monitors, a parameterized model is developed for the purposes of classification -instead of estimating the model parameters, however, PAIN classification provides guaranteed performance (i.e., constant false positive rates) regardless of the parameter values.…”

“…In PAIN monitors, a parameterized model is developed for the purposes of classification -instead of estimating the model parameters, however, PAIN classification provides guaranteed performance (i.e., constant false positive rates) regardless of the parameter values. This approach has been successfully applied in multiple CPS domains with parameterized models such as fault detection in networked systems [79], [80] as well as the development of heating, ventilating and air conditioning systems [86], smart grids [52], and in medical monitoring applications for critical pulmonary shunts [81], [82], hypovolemia [83] and meal-detection in type I diabetics [84], [85].…”

Context-Aware Detection in Medical Cyber-Physical Systems

“…The large amounts of digital data collected by these devices [7], [16] provide great opportunities for developing Medical Cyber-Physical Systems (MCPS) in order to improve health outcomes and reduce costs [20]. Such systems would aid clinicians in multiple ways, ranging from providing prompts to clinicians (in case they are focused on the patient and not looking at the monitors), to alerting clinicians of unsuspected events (by processing time-series data and discovering trends over a long period of time, e.g., in pulmonary shunt detection [12], [13] and hypovolemia detection [22]), to providing fully closed-loop systems (e.g., the artificial pancreas project [6]).…”